| //! miranda-nodes — Work Order 3: the 52-channel ARKit blendshape SIMD | |
| //! solver, Perlin-noise micro-saccades, breathing oscillators, and | |
| //! blendshape velocity clamping. | |
| //! | |
| //! # Two speech sources, one face | |
| //! | |
| //! Speech-driven mouth shapes come from either of two interchangeable | |
| //! sources, chosen per pipeline: | |
| //! | |
| //! - **Pipeline 1 (cloud)**: [`viseme`] — Amazon Polly viseme events | |
| //! mapped through an interpolation table. | |
| //! - **Pipeline 2 (local)**: the SIMD acoustic-energy solver (WO-3 T7), | |
| //! computing weights straight from PCM features. | |
| //! | |
| //! Everything downstream of that choice is shared: the autonomic layer | |
| //! (blink, gaze, breath) and the compositor/damper run identically | |
| //! regardless of which source produced the speech weights. That's what | |
| //! makes the speech source a swappable role slot rather than a fork in | |
| //! the whole face pipeline. | |
| //! | |
| //! # The autonomic layer is always running | |
| //! | |
| //! Per the Instant Presence Standard's No-Loop Video Protocol | |
| //! (`eve-ecc-docs/INSTANT-PRESENCE-STANDARD.md`), the oscillators run | |
| //! continuously whether or not anyone is speaking. "Idle" and "alive" are | |
| //! the same code path here — there is deliberately no separate idle mode | |
| //! that a canned loop could ever be substituted into. | |
| pub mod blink; | |
| pub mod breath; | |
| pub mod compositor; | |
| pub mod conversation; | |
| pub mod dispatcher; | |
| pub mod forge; | |
| pub mod gaze; | |
| pub mod memory; | |
| pub mod solver; | |
| pub mod verify; | |
| pub mod viseme; | |